Les 3 et 4 avril 2026 apportent deux décisions majeures côté Anthropic : un changement de politique qui retire l’accès via outils tiers des abonnements Claude Pro et Max, accompagné de la publication de Claude Code v2.1.92. Du côté de Google, trois nouvelles fonctionnalités sont annoncées : le modèle vidéo Veo 3.1 Lite, l’intégration de Lyria 3 dans l’app Gemini, et la génération vidéo haute qualité dans Google Workspace.
Third-party tools excluded from Claude Pro and Max subscriptions
April 3 — Boris Cherny (PM Claude Code at Anthropic) announced on X that Claude subscriptions (Pro, Max) will no longer cover usage via third-party tools that access Claude through the API.
These tools — which were bypassing intended usage limits for direct usage — created pressure on available capacity for all subscribers. The change aims to restore service quality for users accessing Claude directly via claude.ai.
We’ve been working hard to meet the increase in demand for Claude, and our subscriptions weren’t built for the usage patterns of these third-party tools. Capacity is a resource we [need to manage]. — @bcherny on X
Support measures for affected subscribers:
| Measure | Details |
|---|---|
| One-time credit | Equivalent to the monthly cost of the subscription |
| Usage bundles | Option to buy API credits at a reduced rate |
| Full refund | Via a link sent by email upon request |
We want to be intentional in managing our growth to continue to serve our customers sustainably long-term. This change is a step toward that. — @bcherny on X
Claude Code v2.1.92: Bedrock wizard, per-model costs, interactive /release-notes
April 4 — Claude Code v2.1.92 was published on npm with several improvements for Enterprise teams and Pro users.
New features
| Feature | Description |
|---|---|
forceRemoteSettingsRefresh | Policy parameter: blocks startup until remote settings are fetched (fail-closed mode) |
| Interactive Bedrock assistant | Bedrock setup wizard from the sign-in screen — guides AWS configuration, region, credential verification, and model selection |
| Per-model cost breakdown | /cost now shows the breakdown by model and cache hits for subscribers |
Interactive /release-notes | Becomes an interactive version selector |
| Remote Control session names | Uses the hostname as the default prefix (e.g. myhost-graceful-unicorn), configurable via --remote-control-session-name-prefix |
| Expired cache indicator (Pro) | Session footer message indicating how many tokens will be sent uncached on the next turn |
Notable fixes
- Fixed an agent crash when tmux windows are removed or renumbered during a session
- Fixed Stop hooks being incorrectly rejected when the fast model returns
ok:false - Fixed 400 API errors related to empty text blocks produced by “extended thinking” mode
- 60% speed improvement for the Write tool diff on large files
- Removed commands
/tagand/vim(vim mode is available via/config)
Veo 3.1 Lite: fast video generation at half price in the API
April 3 — Google launched Veo 3.1 Lite, a new video-generation model available via the Gemini API in Google AI Studio.
| Aspect | Veo 3.1 Lite | Veo 3.1 Fast |
|---|---|---|
| Speed | Same | Reference |
| Cost | Half the price | Reference |
| Availability | API Gemini / AI Studio | API Gemini / AI Studio |
This new pricing tier broadens options for developers with high-volume use cases where generation speed matters but per-request budget must be controlled.
Google Vids: Veo 3.1 video generation for all Workspace users
April 3 — High-quality video generation via Veo 3.1 is now available in Google Vids for all Google Workspace users at no additional cost.
Google Vids is the video-creation tool integrated into Workspace. This integration allows generating presentation, training, or internal-communication videos directly within the professional environment, using the same Veo 3.1 model accessible via the API.
Lyria 3 in the Gemini app: multimodal music generation
April 3 — Google highlights the capabilities of Lyria 3, its music-generation model, now available in the Gemini app.
Users can compose custom tracks from three input types:
| Input mode | Description |
|---|---|
| Text | Prompt describing the style, tempo, instruments |
| Photo | An image as visual inspiration |
| Video | A video clip as reference |
Combining text, photo, and video for music composition positions Lyria 3 as a music/visual synchronization tool, useful for creators who want to adapt music to existing content.
In brief
Computer use on Windows — April 2 — The Computer use feature (control of the computer by Claude) is now available on Windows in Claude Cowork and Claude Code Desktop, complementing availability already provided on macOS.
Focus mode in Google AI Studio — April 3 — Google AI Studio receives a Focus mode allowing targeted editing of specific sections of an AI-generated app without recompiling the entire project.
Crosscoders — inter-architecture model comparison — April 3 — Anthropic Fellows published research on the Cross-Architecture Model Diffing method with Crosscoders: applying the software “diff” principle to analyze internal representations of open-weight models to identify behaviors specific to each architecture.
🔗 arXiv paper | Announcement @AnthropicAI
What this means
Anthropic’s subscription decision is the most directly impactful for users. It reveals a structural tension: third-party tools had found a cost-effective way to access Claude via consumer subscriptions, which wasn’t anticipated in the pricing model. The compensatory measures (credit, bundles, refunds) show that Anthropic does not want to block these usages but to have them contribute to the real cost via the API.
Claude Code v2.1.92 continues the Enterprise direction: fail-closed mode for remote settings and the interactive Bedrock wizard reduce friction for teams deploying Claude Code in AWS environments. The expired cache indicator for Pro subscribers is a useful transparency improvement to help control costs.
On Google’s side, launching Veo 3.1 Lite in the API and integrating Veo 3.1 into Workspace follow the same democratization logic: offer multiple cost/quality tiers to broaden adoption. Lyria 3 in Gemini adds a multimodal dimension for music creation that complements the video capabilities announced in parallel.
Sources
- @bcherny on X — subscription policy
- Claude Code v2.1.92 CHANGELOG
- @claudeai on X — Computer use Windows
- @AnthropicAI on X — Crosscoders
- Crosscoders — arXiv
- @GoogleAI on X — Veo 3.1 Lite, Google Vids, Focus mode
- @Google on X — Lyria 3 in Gemini
This document was translated from the French version into English using the gpt-5-mini model. For more information about the translation process, consult https://gitlab.com/jls42/ai-powered-markdown-translator